Add a new function TMath::Power(double, int) impelmented using std::pow(double,int) which is 100% faster than the double version

New KDTree class from C. Gumpert. which has different splitting rules using the data population. The splitting can is based on the basket population, but any function of the data can be used. For example, in case of weighted data one can split according to the basket total weight or the effective entries in the basket. In this way the class can support weighted data sets. The splitting of the TKDTree class is instead fixed and based on the basket number of entries.
The tree can also be frozen to fix the splitting and behaving like a multi-dim histograms with bins with variables hyper-volumes.
Auxiliary classes are provided like the node classes or the data point class, which can exists also with a compile timed fixed dimension for better performances.

Provide support for weighted likelihood unbinned fits
Provide support for extended likelihood unbinned fits
Provide support for not-extended binned likelihood fits (i.e. multinomial distribution instead of Poisson for bin contents)
In case of binned likelihood fit build a Poisson or (Multinomial) likelihood ratio with the saturated model. So a constant term is subtracted to the likelihood. The formulae described in the Baker and Cousins paper (N.I.M. 221 (1984) 437) are now used. The obtained negative likelihood ratio value from the fit and multiplied by a factor 2 is now asymptotically distributed as a chi square.